Since upgrading from Jetpack 6.1 and Deepstream 7.1, to Jetpack 7.2 and Deepstream 9 on my jetson orin nano, I am unable to run custom and generic YOLO models in Deepstream. Inference with NVIDIA’s stock INT8 models work, but I haven’t been able to get any YOLO models to run inference and I did not have an issue before upgrading.
Am I encountering this problem because Deepstream 9 is not yet compatible with with the orin series? Or am I just doing something else wrong?
*** Please note that this reply is generated by LLM automatically ***
The issue you’re experiencing with running custom and generic YOLO models in Deepstream Inference on your Jetson Orin Nano after upgrading from Jetpack 6.1 and Deepstream 7.1 to Jetpack 7.2 and Deepstream 9 is likely due to compatibility issues.
Deepstream 9 is compatible with the Ampere, Jetson, Hopper, Lovelace, Pascal, and Turing architectures, but there might be specific requirements or limitations for the Orin series.
Here are a few potential reasons why you’re encountering this issue:
Model compatibility: Deepstream 9 might have changed the way it handles YOLO models, and your custom models might not be compatible with the new version. You can try checking the Deepstream documentation for any changes in model support or requirements.
INT8 calibration: Since NVIDIA’s stock INT8 models work, it’s possible that the issue lies in the INT8 calibration process for your custom YOLO models. You can try recalibrating your models using the Deepstream calibration tools to ensure they’re compatible with the Orin series.
Orin series specific requirements: The Orin series might have specific requirements or limitations for running YOLO models in Deepstream Inference. You can try checking the Jetson Orin NX Series Modules Data Sheet and the Deepstream documentation for any Orin-specific requirements or recommendations.
Deepstream configuration: It’s possible that the Deepstream configuration has changed between versions, and you might need to update your configuration files to work with Deepstream 9. You can try checking the Deepstream documentation for any changes in configuration requirements.
To troubleshoot the issue, you can try the following:
Check the Deepstream documentation: Review the Deepstream documentation for any changes in model support, INT8 calibration, or configuration requirements.
Verify model compatibility: Check if your custom YOLO models are compatible with Deepstream 9 and the Orin series.
Recalibrate INT8 models: Try recalibrating your custom YOLO models using the Deepstream calibration tools to ensure they’re compatible with the Orin series.
Update Deepstream configuration: Update your Deepstream configuration files to work with Deepstream 9, if necessary.
Contact NVIDIA support: If none of the above steps resolve the issue, you can contact NVIDIA support for further assistance.
In comparison to the previous version, Deepstream 9 might have introduced changes that affect the way YOLO models are handled. By checking the documentation and verifying model compatibility, you should be able to identify the root cause of the issue and find a solution.
Here’s a summary of the key points:
Deepstream 9 is compatible with the Ampere, Jetson, Hopper, Lovelace, Pascal, and Turing architectures.
The Orin series might have specific requirements or limitations for running YOLO models in Deepstream Inference.
Model compatibility, INT8 calibration, and Deepstream configuration might be the root cause of the issue.
Troubleshooting steps include checking the Deepstream documentation, verifying model compatibility, recalibrating INT8 models, updating Deepstream configuration, and contacting NVIDIA support if necessary.
*** Please give the thumbs up if you get answers you like. Or provide feedback to help us improve the answer. ***
DS-next is fully compatible with JP-7.2 and will be released in the future; DS-9.0 has not been fully tested with JP-7.2 and may encounter some compatibility issues.
But when I wanted to make the last step and package it in the Deepstream-9.0 Container I noticed, that the official container was running on older trt and CUDA versions, than the Jetson itself. Turns out, the older version is not compatible with the Orins Hardware.
So guess I have to wait for the Next Deepstream Version being JP-7.2 compatible?